Public Detailed service description
Mahti has a total of 1404 CPU nodes and 24 GPU nodes. The theoretical peak performance is 7.5 petaflops for the CPU nodes and 2.0 petaflops for the GPU nodes, in total 9.5 petaflops.
Both CPU and GPU nodes have two AMD Rome 7H12 CPUs with 64 cores each, making the total core count about 180 000. The CPUs are based on AMD Zen 2 architecture, supporting the AVX2 vector instruction set, and run at 2.6 GHz base frequency.
The 1404 CPU nodes are equipped with 256 GB of memory and the vast majority have no local disks. There are in total 60 nodes that are equipped with a local 3.8 TB NVMe drive. These are available in the small
and interactive
partitions.
The 24 GPU nodes are equipped with 512 GB of memory, a local 3.8 TB NVMe drive and four Nvidia Ampere A100 GPUs. In a subset of the nodes the A100 GPUs have been split into multiple smaller GPUs with a fraction of the compute and memory capacity of the A100 GPUs. These are useful for interactive work, courses and for code development.
Jobs are submitted via batch queueing system. Small jobs, as well as medium and very large sized jobs can be run on Mahti.
The main CPU partitions provide full nodes, and a maximum job duration of 36 hours. The maximum job size is 20 nodes, but this is increased to 200 nodes if the scalability of the jobs is demonstrated. There are also smaller partitions providing core-based allocations on shared nodes. Here the jobs can run up to 7 days. The GPU partition provides access to up to 6 nodes or 24 GPUs per job. The maximum job duration is 36 hours.
Access is provided through SSH or web interface using Open OnDemand.
Mahti supports user-developed codes (Fortran, C/C++, Python), Linux-compatible applications via compilation or Singularity containers, and CSC’s scientific software collection. Parallelization options include MPI, OpenMP, CUDA, and OpenACC. Mathematical subroutine libraries are available.
Certifications
The Information Security Management System of CSC – IT Center of Science Oy is ISO/IEC 27001:2022 certified, with the Mahti service being within scope of the certificate.
User content including personal data in the Service
User must evaluate whether this Service is suitable for their purpose. Please pay attention to special categories of personal data.
When CSC processes personal data in CSC’s services on behalf of the controller, we always require a personal data processing agreement. The user creates the description of processing activities in MyCSC customer portal. Data Processing Agreement.
The Service do not assert ownership or any intellectual property rights to users or customers organisations’ content in the services.
Client’s responsibilities
User must follow all the Terms of Use that applies using the Service
Users are also responsible for their data and computing and should always strive to use the provided resources efficiently. The user is responsible to minimize the chance of overloading any component of the service by excessive amount or frequency of operations. CSC has no backups and backups are the responsibility of users. Users are responsible for adhering to the compute services usage policies.
Service producer’s responsibilities
The service producer ensures that the service is available to customers as described in the service description.
The service producer is responsible for producing and developing the Service.
Service producer
CSC – IT Center for Science Ltd.
Service provider
CSC – IT Center for Science Ltd.
Last updated
2025-06-13